Different aspects of a sentence may contain different sentiments, and sentiment descriptors for a given aspect exist in different places in the sentence, making it difficult to determine its sentiment polarity. Aiming at the above problems, a neighborhood network (Nenet) for aspect-based sentiment analysis is proposed. Firstly, the context information of the text is encoded, and the neighborhood information of the target aspect at the grammar level is extracted by using a graph convolutional neural network. At the same time, the convolutional neural network is used to extract the neighborhood information at the physical level, and the two extracted features are combined to improve the text expression ability. Finally, an attention mechanism is used to express the key information in sentences for judging sentiment polarity, and the final text representation is input to the sentiment analysis layer to predict sentiment polarity. Experiments are carried out on three standard datasets, and the experimental results show that the neighborhood network outperforms other baseline models in both index accuracy and F1 value.